[Docs] How I Built MMCV 2.1.0 on Windows 11 for RTX 5060 Ti (CUDA 12.8)

[semihhdemirel]

📚 The doc issue

Build MMCV 2.1.0 from Source on Windows (RTX 50 Series / CUDA 12.8)

This guide documents how I successfully built MMCV from source on Windows for an RTX 5060 Ti.

Environment

Component	Version
OS	Windows 11
Python	3.10.20
CUDA Toolkit	12.8
PyTorch	2.7.1+cu128
GPU	NVIDIA RTX 5060 Ti 16GB
Visual Studio	Build Tools 2022
MSVC	v142 (14.29)
Ninja	latest

1. Install Visual Studio Build Tools

Make sure these are installed:

• MSVC v142 - VS 2019 C++ x64/x86 build tools
• Windows 10/11 SDK
• C++ CMake tools
• Ninja

2. Clone MMCV

git clone https://github.com/open-mmlab/mmcv.git
cd mmcv

git checkout v2.1.0

# 3. Install Python dependencies

```bash
pip install -r requirements/runtime.txt
pip install -r requirements/build.txt

pip install "setuptools<81" wheel

---

4. Open the correct Developer Command Prompt

Use the x64 Native Tools Command Prompt or run:

call "C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Auxiliary\Build\vcvarsall.bat" x64 -vcvars_ver=14.29

Verify:

cl

Expected:

Microsoft (R) C/C++ Optimizing Compiler Version 19.29.xxxxx

---

5. Activate your conda environment

conda activate mmdetection

---

6. Set environment variables

Command Prompt:

set CUDA_HOME=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.8
set MMCV_WITH_OPS=1
set MAX_JOBS=4
set DISTUTILS_USE_SDK=1
set TORCH_CUDA_ARCH_LIST=12.0

PowerShell equivalent:

$env:CUDA_HOME="C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.8"
$env:MMCV_WITH_OPS="1"
$env:MAX_JOBS="4"
$env:DISTUTILS_USE_SDK="1"
$env:TORCH_CUDA_ARCH_LIST="12.0"

---

7. Build MMCV

pip install . -v --no-build-isolation

The build takes several minutes.

---

8. Verify installation

import mmcv
import mmcv._ext

print(mmcv.__version__)

Suggest a potential alternative/fix

No response